Abstract

Interaction among autonomous agents in Multi-Agent Systems (MASs) is the key aspect for solving complex problems that an individual agent cannot handle alone. In this context, social approaches, as opposed to the mental approaches, have recently received a considerable attention in the area of agent communication. They exploit observable social
commitments to develop a verifiable formal semantics by which communication protocols can be specified. However, existing approaches for defining social commitments tend to
assume an absolute guarantee of correctness so that systems run in a certain manner. That is, social commitments have always been modeled with the assumption of certainty. Moreover, the widespread use of MASs increases the interest to explore the interactions between different aspects of the participating agents such as the interaction between agents’ knowledge and social commitments in the presence of uncertainty. This results in having a gap, in the literature of agent communication, on modeling and verifying social commitments in probabilistic settings.

In this thesis, we aim to address the above-mentioned problems by presenting a practical formal framework that is capable of handling the problem of uncertainty in social
commitments. First, we develop an approach for representing, reasoning about, and verifying
probabilistic social commitments in MASs. This includes defining a new logic called the probabilistic logic of commitments (PCTLC), and a reduction-based model checking
procedure for verifying the proposed logic. In the reduction technique, the problem of model checking PCTLC is transformed into the problem of model checking PCTL so that
the use of the PRISM (Probabilistic Symbolic Model Checker) is made possible. Formulae of PCTLC are interpreted over an extended version of the probabilistic interpreted systems
formalism. Second, we extend the work we proposed for probabilistic social commitments to be able to capture and verify the interactions between knowledge and commitments.
Properties representing the interactions between the two aspects are expressed in a new developed logic called the probabilistic logic of knowledge and commitment (PCTLkc).
Third, we develop an adequate semantics for the group social commitments, for the first time in the literature, and integrate it into the framework. We then introduce an improved version of PCTLkc and extend it with operators for the group knowledge and group social commitments. The new refined logic is called PCTLkc+. In each of the latter stages, we respectively develop a new version of the probabilistic interpreted systems over which the
presented logic is interpreted, and introduce a new reduction-based verification technique to verify the proposed logic. To evaluate our proposed work, we implement the proposed verification techniques on top of the PRISM model checker and apply them on several case studies. The results demonstrate the usefulness and effectiveness of our proposed work.